Assessment of Transition Model and CFD

Methodology for Wind Turbine Flows

Aniket C. Aranake

Vinod K. Lakshminarayan

Karthik Duraisamy

A detailed evaluation of the predictive capability of a Reynolds Averaged Navier Stokes
(RANS) solver with a transition model is performed for wind turbine applications. The
performance of the computational methodology is investigated in situations involving at-
tached ow as well as incipient and massive ow separation and compared with experiment.
Two-dimensional simulations on wind turbine airfoil sections are seen to qualitatively and
quantitatively predict the onset of transition to turbulence and provide signicantly im-
proved lift and drag predictions when compared to simulations that assume fully turbulent
ow. In three-dimensional wind turbine simulations, detailed validation studies of the in-
tegrated loads and sectional pressure coecient also show denite improvements at wind
speeds at which separation is incipient or conned to a small portion of the blade sur-
face. At low wind speeds, for which the ow is mostly attached to the blade surface, and
at high wind speeds, for which it is massively separated, the transition model produces
similar results to a fully turbulent calculation. Overall, the performance of the transition
model highlights the necessity of such models while also pointing out the need for further
development.
Nomenclature
Angle of attack
Intermittancy

T
Freestream eddy viscosity
Specic dissipation
Re
t
Transition momentum thickness Reynolds number
Blade geometric twist
Density of air
c Chord length
c
d
Sectional drag coecient
c
f
Skin friction coecient
c
l
Sectional lift coecient
C
p
Pressure coecient
C
Q
Torque coecient
C
T
Thrust coecient
k Turbulent kinetic energy
M
FS
Freestream Mach number
M
tip
Rotor tip Mach number
P Static pressure
P

Freestream static pressure

Q Torque
R Rotor radius
r Rotor spanwise coordinate
Re

Momentum thickness Reynolds number

PhD Candidate, Department of Aeronautics and Astronautics, Stanford University, Stanford, CA.

Postdoctoral Fellow, Department of Aeronautics and Astronautics, Stanford University, Stanford, CA.

Assistant Professor (Consulting), Department of Aeronautics and Astronautics, Stanford University, Stanford, CA.
1 of 22
American Institute of Aeronautics and Astronautics
42nd AIAA Fluid Dynamics Conference and Exhibit
25 - 28 June 2012, New Orleans, Louisiana
AIAA 2012-2720
Copyright 2012 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
Re
x
Local position Reynolds number
T Thrust
Tu

Freestream turbulence intensity

V

Freestream velocity
V
tip
Rotor tip velocity
x
trans
Chordwise transition location
I. Introduction
In recent decades, wind energys share of the global energy portfolio has been increasing, a trend which
is expected to continue as fossil fuels are replaced by renewable alternatives. In the U.S., the Department
of Energy has published a report outlining its plan to satisfy 20% of the national energy budget with wind
power.
1
In a best case scenario, the report estimates that annual wind energy production can increase by 35%
due to improvements related to aerodynamics. As technology matures, achieving these gains in performance
will depend on an increasingly sophisticated understanding of physics, and thus classical engineering models
reach their limit of usefulness. Higher delity methods, such as the one discussed in the present work,
provide designers with the ability to assess the performance of their designs with high condence. This
information reduces uncertainty for businesses and promotes the development of wind energy solutions.
Additionally, advanced concept designs such as diuser augmented wind turbines (DAWTs), active ow
control, and plasma actuation rely on nonlinear and interactional aerodynamic eects, so the adoption of
these technologies is increasingly reliant on the availability of fast and reliable modeling tools.
The simplest models for wind turbine aerodynamic analysis use momentum theory and blade element
theory which make use of the integral mass, momentum, and energy equations in a control volume around
the turbine along with simple aerodynamic models or tabulated airfoil data. While such models provide
some practical insight and sizing estimates, they are of limited use in the design of modern wind turbines.
An introduction to low-order wind turbine aerodynamics including details of these methods can be found in
Hansen.
2
More sophisticated techniques to include the eects of the evolving wake system of turbines and
to include unsteady gusts or turbine yaw can be derived using vortex panel/lament based methods.
3
In all of these methods, rst order eects such as viscosity and turbulence can only be treated in a
purely empirical manner. Physically accounting for these eects requires higher delity CFD modeling. An
assessment of predictive capabilities by Simms et al.
4
revealed that dierent models produce widely varying
results. Experimental data from this study, the NREL Phase VI Rotor Unsteady Aerodynamics Experiment,
was published
5
and has since been used to assess wind turbine aerodynamic models in several publications
(see, for example, Srensen et al.,
6
Duque et al.,
7
Potsdam et al.
8
)
A principal diculty in the accurate simulation of wind turbine aerodynamics is an accurate represen-
tation of separated ow. In this regard, predicting the transition from laminar to turbulent ow may be
extremely important, especially when laminar separation bubbles (LSB) are present near the leading edge
of the airfoil. In such cases, if transition occurs over the bubble, the LSB may burst, causing leading edge
stall. This process may be highly sensitive to even low energy perturbations and phenomena such as the
double stall phenomenon can occur.
9
A classic approach to predicting transition is the method of Michel,
10
which triggers transition based on an empirical correlation between momentum thickness Reynolds number
Re

and Reynolds number based on local length Re

x
at the point of transition. This model is suciently
accurate for many airfoil-type problems,
11
but does not account for surface roughness. For wind turbine
airfoils, a commonly used technique is that of Eppler,
12
which has been employed in the design of several
NREL S-series wind turbine airfoils.
13, 14
This method, which is also empirical, incorporates a roughness
factor into a comparison between integral boundary layer quantities to decide when to trigger transition. For
practical problems, these methods typically involve a numerical integration of the boundary layer followed
by a search for critical momentum Reynolds number, both of which can be expensive and dicult to apply
in three dimensions.
An alternate approach to identifying transition is through the use of stability analysis. The Orr-
Sommereld equations,
15
based on the linearized Navier-Stokes equations, describe when the linearized
modes in parallel shear ows will amplify, causing the ow to become turbulent. This simplied view of the
transition phenomenon neglects nonlinear eects, which can create large destabilizing transients. A more
commonly used method is the e
n
method of Smith et al.
16
and Van-Ingen.
17
These methods correlate the
amplitude ratio of the largest mode to the location of transition. A shortcoming of this type of method is
2 of 22
American Institute of Aeronautics and Astronautics
that it requires the computation of streamlines, adding signicantly to the cost and complexity of a CFD
simulation. An extensive review of transition models for CFD has been published by Pasquale et al.
18
Direct methods for the prediction of transition and separated ow such as DNS or LES are frequently
prohibitively expensive for industrial applications. For this reason, a transition model augmenting the RANS
equations is desirable. To further avoid the diculties of the methods mentioned above, it is desirable to
select a method that makes use of local quantities. With this property, a model can be readily implemented
into an existing RANS solver. A recently published model that satises this criteria is the Re
t
model
by Langtry et al.
19
This model requires the solution of two transport equations in addition to the RANS
equations and the SST k turbulence model.
21
An investigation by the same authors
20
found that this
model improves signicantly the prediction of torque in wind turbine ows involving separation. However,
the quality of the results is somewhat obscured by the discrepancy of the predictions with measured results
in the attached ow regime. Since its development, the Re
t
model has been adapted by Medida et
al.
22
for use with the Spalart-Allmaras turbulence model.
23
Compared to popular two equation turbulence
models, the Spalart-Allmaras model is more widely used in applications involving external ows, and it is
also typically less expensive and more numerically well-behaved.
The objective of the current work is to carefully evaluate the applicability of RANS-based ow solver and
Re
t
SA model for wind turbine simulations. Specic emphasis is given to transition to turbulence
caused by the Laminar separation bubble. Simulations with and without the use of this transition model are
performed in two dimensional airfoil ows as well as in wind turbine rotors. For simulations run without the
transition model, the ow is assumed to be fully turbulent and the Spalart-Allmaras method is employed
in its original form. The results are carefully validated against experimental data, and the predicted ow
physics are analyzed in detail.
II. Flow Solver
In this work, computations are performed using the overset structured mesh solver OVERTURNS.
24
This
code solves the compressible RANS equations using a preconditioned dual-time scheme in the diagonalized
approximate factorization framework, described by Buelow et al.
25
and Pandya et al.
26
The diagonal form
of the implicit approximate factorization method was originally developed by Pulliam and Chaussee.
27
The
low Mach preconditioning is based on the one developed by Turkel.
28
The preconditioning is used not only
to improve convergence but also to improve the accuracy of the spatial discretization. The inviscid terms are
computed using a third order MUSCL scheme utilizing Korens limiter with Roes ux dierence splitting
and the viscous terms are computed using second order central dierencing. For RANS closure, the Spalart-
Allmaras
23
turbulence model is employed. To the eects of ow transition, the RE
t
SA
22
model is
used. For completeness, details of the model is included in the appendix.
III. 2D Validation Results
For an initial assessment of the eectiveness of the Re
t
SA transition model, two-dimensional
simulations are performed on two dierent NREL S-series airfoils, the S827 and the S809. These airfoils are
widely used in wind turbines, and their physics is strongly characterized by the eects of transition.
Experimental results
13, 14
are available for both airfoils at dierent angles of attack at a freestream Mach
number of 0.1. The Reynolds number for the S827 airfoil is 3 10
6
and that for the S809 airfoil is 2 10
6
.
The freestream turbulence intensity is less than 0.05%. C-O type meshes are used for the simulations. The
baseline grid has 527 101 points in the wrap-around and normal directions, respectively. Figure 1 shows
the grid used for S827 calculation.
A. S827 Airfoil
The S827 airfoil, used in stall-regulated wind turbines, is designed to deliberately induce ow separation
beyond a prescribed angle of attack. The separated region is conned to a separation ramp
14
near the
trailing edge of the upper surface so that the airfoil may continue to operate with reduced lift. Accordingly,
a salient feature of the S827 lift curve is its two distinct slopes, one corresponding to angles below stall and
a second slope for the post-stall region beginning at and angle of attack around = 5

. It is particularly
dicult to capture this double slope behavior using RANS-based simulations, and therefore this case serves
3 of 22
American Institute of Aeronautics and Astronautics
Figure 1. Baseline grid for S827 calculation
as a good test for the transition model.
Figure 2 shows the mean lift coecient as a function of angle of attack along with error bars representing
the range of unsteady variation in the time-accurate simulations. In Fig. 2(a), the results from the simulations
with and without the transition model are compared to the experimental data. The eectiveness of the
transition model is evident from the plot. While the fully turbulent simulation fails to capture the double
slope and predicts a much higher maximum lift coecient as compared to the experimental value, the
simulation using the transition model captures both the lift slopes as well. In addition, the transition model
captures the maximum lift coecient accurately. The only discrepancy is that the predicted angle at which
the lift slope changes is 7

instead of 5

. Figure 2(b) shows the results obtained from a sensitivity study

on grid renement. The results are computed on a ne 727 101 grid and a coarse 327 85 grid. Here,
the baseline grid is referred to as the medium grid. The computed lift coecients change only slightly
between the grids, giving further condence in the results. A more detailed validation of the transition
model is obtained by comparing the surface pressure coecient distribution. Shown in Fig. 3 are the results
at two angles of attack, 4

and 12

. As with the integrated lift coecient, the use of the transition model
signicantly improves the quality of the solution at both angles of attack.

c
l
0 4 8 12 16 20 24
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
SA + Transition
SA, Fully Turbulent
Experiment
(a) With and without transition model

c
l
0 4 8 12 16 20 24
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Coarse
Medium
Fine
Experiment
(b) Varying grid renement
Figure 2. Lift coecient vs angle of attack for S827 airfoil.
4 of 22
American Institute of Aeronautics and Astronautics
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-1.5
-1
-0.5
0
0.5
1
1.5
SA + Transition
SA, Fully Turbulent
Experiment
(a) = 4

x/c
C
p
0 0.2 0.4 0.6 0.8 1
-8
-6
-4
-2
0
(b) = 12

Figure 3. Pressure coecient distributions for S827 airfoil.

To further illustrate the dierences between solutions obtained from fully turbulent simulations and those
using the transition model, the eddy viscosity contours at 4

and 12

angles of attack are shown in Figs. 4

and 5 respectively. At 4

angle of attack, where the ow is laminar in the front half of the airfoil, the
simulation using the transition model has very low levels of eddy viscosity until the transition ramp near
mid-chord. The eddy viscosity then increases in the turbulent rear half of the airfoil. On the other hand, in
the fully turbulent simulation the eddy viscosity begins growing in magnitude immediately from the leading
edge of the airfoil, resulting in higher overall levels. At 12

angle of attack, transition is expected to occur at

the leading edge of the airfoil and to separate on the aft side of the airfoil. At this angle, the eddy viscosity
levels start growing at the leading edge in both the simulations, however the use of the transition model
(a) With transition model
(b) Fully turbulent
Figure 4. Contours of eddy viscosity at = 4

for S827 airfoil.

5 of 22
American Institute of Aeronautics and Astronautics
(a) With transition model
(b) Fully turbulent
Figure 5. Contours of eddy viscosity at = 12

for S827 airfoil.

(a) = 4

(b) = 12

Figure 6. Contours of intermittency with streamlines for S827 airfoil.

6 of 22
American Institute of Aeronautics and Astronautics

c
l
0 4 8 12 16 20 24
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Tu=0.025
Tu=0.05
Tu=0.1
Experiment
(a) Variation with freestream turbulence intensity

c
l
0 4 8 12 16 20 24
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
=0.05
=0.1
=0.2
Experiment
(b) Variation with freestream eddy viscosity
Figure 7. Lift coecient vs angle of attack for S827 airfoil.
shows a lower growth rate. As a result, the transition model predicts lower viscous damping which leads to
increased separation in the aft of airfoil. When the ow separates, the simulation using the transition model
shows higher levels of eddy viscosity.
The behavior of the transition model is further exemplied by contours of intermittency, which are shown
along with velocity streamlines in Fig. 6. Intermittency is a factor that directly aects the production of
eddy viscosity; values less than one indicate decreased production, while a value of one restores production
to that of the fully turbulent Spalart-Allmaras model. At both angles of attack, the intermittency values
are below one near the airfoil surface, but when the ow separates at 12

angle of attack, the intermittency

increases rapidly. The velocity streamlines indicate the presence of a small separation bubble at the 67%
chord location at 4

angle of attack, as well as the large separated region on the upper surface at 12

angle
of attack.
It is also critical to understand the sensitivities of the ow solution to freestream turbulence intensity
(Tu

) and freestream eddy viscosity (

T
), since these quantities can be expected to vary signicantly in an
actual wind turbine environment. Figures 7(a) and 7(b) show the eect of perturbations to these parameters
on lift coecient. The lift is found to be reasonably insensitive to these parameters for the chosen range of
values.
B. S809 Airfoil
The S809 airfoil is used in the NREL Phase VI wind turbine, against which the 3D solver is validated. For
this reason, a 2D validation study is performed on this airfoil prior to 3D simulations. Similar to the S827
airfoil, the downstream half of the top surface of this airfoil is designed to have a transition ramp,
13
an
extended region of gentle pressure recovery which encourages the smooth transition of ow from laminar to
turbulent.
Results for lift coecient and drag coecient of the S809 airfoil are shown in Figure 8. Again the
Re
t
SA model signicantly improves the prediction of both quantities, particularly in the post-stall
region. The use of the transition model is also seen to predict the pressure coecient accurately (Fig. 9).
Notably, the transition model correctly captures the dip in pressure distribution which indicates the location
of transition.
The transition locations for each angle of attack are plotted in Figure 10. Transition is triggered by
laminar separation bubbles, which span several grid points in the CFD solutions. Error bars are used to
represent the location and width of these bubbles. The transition locations are captured well except in the
range between = 6

and = 8

on the upper surface of the airfoil. The calculations predict that the
separation bubble on the upper surface jumps to the leading edge of the airfoil at = 9

. The ow near the

surface was found to be volatile in the middle range of angles and highly sensitive to small disturbances.
7 of 22
American Institute of Aeronautics and Astronautics

c
l
0 4 8 12 16 20 24
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
(a) c
l
vs

c
d
0 4 8 12 16 20 24
-0.1
0
0.1
0.2
0.3
0.4
SA + Transition
SA, Fully Turbulent
Experiment
(b) c
d
vs
Figure 8. Lift and drag coecients vs angle of attack for S809 airfoil.
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-1.5
-1
-0.5
0
0.5
1
1.5
SA + Transition
SA, Fully Turbulent
Experiment
(a) = 4

x/c
C
p
0 0.2 0.4 0.6 0.8 1
-10
-8
-6
-4
-2
0
2
(b) = 12

Figure 9. Pressure coecient distribution for S809 airfoil.

IV. 3D Validation Results
The performance of the Re
t
SA transition model and three dimensional solver are assessed against
experimental data available from the NREL Phase VI Unsteady Aerodynamics Experiment (UAE).
5
The
conguration is a two-bladed wind turbine consisting of S809 airfoil sections. The blade radius is R = 5.029 m
and the chord length is tapered from c = 0.737 m to c = 0.305 m. The blade twist varies from = 20

to
= 2.5

from root to tip. A detailed geometric description, including the distributions of twist and taper
at spanwise locations, is provided in the UAE report. The blade is rotated at an RPM of 72 and the tip
Reynolds number is 2.42 10
6
. Data is available for wind speeds ranging from 7 m/s to 25 m/s
The computations are performed on an overset mesh system consisting of a C-O type blade mesh and
a background mesh. To reduce computational cost, the simulations are performed in a half-cylinder back-
ground mesh with periodic boundary conditions to represent a cylindrical domain. Information is exchanged
between the meshes using Chimera interpolations. An implicit hole-cutting technique developed by Lee
29
and improved by current authors
24
is used to determine the connectivity information between various over-
set meshes. The blade mesh has 257 51 51 points in the wrap-around, spanwise and normal directions,
8 of 22
American Institute of Aeronautics and Astronautics

x
t
r
a
n
s
0 4 8 12 16
0
0.2
0.4
0.6
0.8
1
1.2
Upper surface
Lower surface
Figure 10. Transition locations for S809 airfoil. The square symbols in this plot are experimental data points.
Error bars show the size of the separation bubble in CFD solution.
respectively. The background mesh has 201 133 164 in the azimuthal, radial, and axial directions, re-
spectively, with azimuthal renement in a 15

patch encompassing the blade. The most rened part of the

background mesh has a grid spacing of 0.02c in the radial and axial directions and an azimuthal spacing of
0.33

. Figure 11 shows the mesh system used, along with a sample ow-eld to demonstrate the quality of
the simulations. Computations were also run on a ner grid with 257 201 101 points for the blade and
180 264 228 points for background mesh for the V

= 7 m/s conguration. The resulting integrated

loads and pressure coecients were conrmed to dier by less than 3%.
Figure 12 compares the predicted integrated thrust and torque to experimental data. As was the case
for 2D simulations, solutions are unsteady, and to account for this variability, the current plot as well as
several subsequent plots in the paper are shown with error bars indicating the range of values attained. At
V

= 7 m/s, where the ow stays attached to the turbine blades, both simulations are able to capture
the integrated quantities equally well. For higher wind speeds, the accuracy of the correlations decreases,
but the overall the trend is well captured. Solutions obtained using the transition model dier from the
fully turbulent simulations only at two middle wind speeds, V

= 10 m/s and V

= 15 m/s. A substantial
X
Y
Z
(a) Blade mesh (b) Sample solution, isocontours of vorticity
Figure 11. Mesh system for NREL wind turbine simulation.
9 of 22
American Institute of Aeronautics and Astronautics
Wind Speed (m/s)
T
h
r
u
s
t
(
N
)
0 5 10 15 20 25 30
0
1000
2000
3000
4000
5000
SA + Transition
SA, Fully Turbulent
Experiment
(a) Thrust
Wind Speed (m/s)
T
o
r
q
u
e
(
N
m
)
0 5 10 15 20 25 30
-500
0
500
1000
1500
2000
2500
(b) Torque
Figure 12. Integrated thrust and torque for NREL wind turbine.
improvement is seen in the prediction of thrust at V

= 10 m/s with the use of the transition model. Torque,

which is more sensitive to small errors in pressure prediction, is not well captured at this speed.
To further investigate the dierences in integrated quantities between the simulations with and without
the transition model, sectional thrust and torque coecients are shown in Fig. 13 for wind speeds of V

=
7 m/s, 10 m/s, and 15 m/s. Thrust and torque coecients are dened by
C
T
=
T
R
2
V
2
tip
(1)
and torque coecient dened by
C
Q
=
Q
R
3
V
2
tip
(2)
where T is the sectional thrust, is freestream density, V
tip
is the tip velocity, and Q is sectional torque.
Unsurprisingly, there is close agreement between the transition model and fully turbulent simulation for
the case at V

= 7 m/s where the ow stays attached. At V

= 10 m/s, the case with the transition

model indicates the presence of ow separation through most of the mid span, while the fully turbulent
simulation does not show this behavior, accounting for the dierences in integrated loads shown earlier. At
V

= 15 m/s, both simulations show decreased loads through most of the span because of ow separation.
The fully turbulent and the transition model results dier substantially in the sectional torque in the outboard
regions, and as a result the fully turbulent simulation predicts a slightly higher integrated torque at this
wind speed. At higher wind speeds (not shown), the proles for the fully turbulent and the transition model
cases show comparable trends.
A more detailed validation of the current methodology is achieved by comparing the predicted pressure
coecient with and without the use of transition model to available experimental data. The pressure
coecient is computed as:
C
p
=
P P

1
2
_
M
2
FS
+
_
M
tip
r
R
_
2
_, (3)
where P is the local static pressure, P

is the pressure in the free stream, M

FS
is the free stream Mach
number, M
tip
is the rotor tip Mach number, R is the rotor radius, and r is the spanwise coordinate at which
the pressure coecient is evaluated.
Pressure coecient distributions at V

= 7 m/s are shown in Fig 14 at four dierent spanwise locations.

At this wind speed, the ow remains attached across the turbine blade, and the results with and without the
use of transition model compare well with the experimental pressure data. The use of the transition model
10 of 22
American Institute of Aeronautics and Astronautics
r/R
R
*
d
C
T
/
d
r
0 0.2 0.4 0.6 0.8 1 1.2
-0.002
0
0.002
0.004
0.006
0.008
0.01
SA + Transition
SA, Fully Turbulent
(a) c
l
at V = 7 m/s
r/R
R
*
d
C
Q
/
d
r
0 0.2 0.4 0.6 0.8 1 1.2
-0.001
-0.0005
0
0.0005
0.001
0.0015
(b) c
d
at V = 7 m/s
r/R
R
*
d
C
T
/
d
r
0 0.2 0.4 0.6 0.8 1 1.2
0
0.005
0.01
(c) c
l
at V = 10 m/s
r/R
R
*
d
C
Q
/
d
r
0 0.2 0.4 0.6 0.8 1 1.2
-0.001
0
0.001
0.002
0.003
(d) c
d
at V = 10 m/s
r/R
R
*
d
C
T
/
d
r
0 0.2 0.4 0.6 0.8 1 1.2
-0.02
-0.015
-0.01
-0.005
0
0.005
(e) c
l
at V = 15 m/s
r/R
R
*
d
C
Q
/
d
r
0 0.2 0.4 0.6 0.8 1 1.2
-0.002
-0.001
0
0.001
0.002
0.003
(f) c
d
at V = 15 m/s
Figure 13. Spanwise load distributions with and without the transition model for NREL wind turbine.
predicts the presence of a separation bubble near the mid-chord region at both upper and lower surface of
the airfoil at all spanwise locations. This leads to a small deviation in the pressure distribution between the
11 of 22
American Institute of Aeronautics and Astronautics
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-4
-3
-2
-1
0
1
2
SA + Transition
SA, Fully Turbulent
Experiment
(a) r/R = 0.3
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-4
-3
-2
-1
0
1
2
(b) r/R = 0.47
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-2
-1.5
-1
-0.5
0
0.5
1
1.5
(c) r/R = 0.80
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-1.5
-1
-0.5
0
0.5
1
1.5
(d) r/R = 0.95
Figure 14. Pressure coecient distributions for NREL wind turbine, V = 7m/s.
two simulations. Although the available experimental data is insucient to conrm the correctness of this
prediction, this feature is very similar to that observed in the S809 airfoil.
Sectional pressure coecients at V

= 10 m/s, where the ow physics is dominated by transitional eects,

are shown in Fig.15. Fully turbulent simulations lead to signicant errors in the prediction of pressure on
the upper surface at 47% and 95% spanwise locations, whereas the transition model correctly captures the
nearly at pressure distribution that suggests separation at the leading edge. However, the simuation with
the transition model shows slightly poorer prediction at 80% spanwise location. Nonetheless, the use of the
transition model leads to an overall improvement at this wind speed.
At V

= 15 m/s, the computational results are nearly identical with and without the transition model
at the inboard span locations shown in Figs. 16(a) and 16(b). At these locations, neither model is able to
capture the pressure well. At the next station, at 80% span, shown in Fig. 16(c), both models match the
data reasonably well. An anomaly appears at the outboard station at 95% span. Here, the transition model
predicts that the ow is separated, whereas the experimental C
p
distribution indicates otherwise. In this
case, the fully turbulent simulation captures the correct pressure.
Results for pressures at V

= 20 m/s and V

= 25 m/s are shown in Figures 17 and 18 respectively.

At these speeds, the ow is completely separated. There is little dierence in performance between the fully
turbulent simulations and those using the transition model, and the results compare reasonably well with
12 of 22
American Institute of Aeronautics and Astronautics
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-10
-8
-6
-4
-2
0
2
SA + Transition
SA, Fully Turbulent
Experiment
(a) r/R = 0.3
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-12
-10
-8
-6
-4
-2
0
2
4
(b) r/R = 0.47
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-6
-4
-2
0
2
(c) r/R = 0.80
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-4
-3
-2
-1
0
1
(d) r/R = 0.95
Figure 15. Pressure coecient distributions for NREL wind turbine, V = 10m/s.
the experimental data.
A more complete picture of the surface pressure is obtained from plots of pressure contours (normalized
by freestream pressure) over the blade surface. Figure 19 compares the surface pressure distribution for
the simulation with and without the transition model at wind speeds of 7, 10, 15 and 20 m/s. Note that
the surfaces are named based on the orientation of the airfoil, which means that the surface referred to as
lower faces the wind and upper faces away. This notation is used consistently in the paper. As observed
before, the most signicant dierence with the use of transition model arises at V

= 10 m/s.
Further examination of the eect of the transition model on the ow-eld is performed by examining skin
friction which is dened as:
c
f
=
2
Re
tip
V
rot
n
1
M
tip
(4)
where Re
tip
is the Reynolds number based on tip velocity, V
rot
is the velocity magnitude in a reference
frame rotating with the blade, and n is the direction normal to the blade surface. Figure 20 shows the skin
friction contours (normalized using tip speed and freestream density) along with velocity streamlines. The
plots are obtained from solution values located one grid cell o the blade surface. At V

= 7 m/s, the
transition model predicts the presence of a separation bubble around the mid-chord region through out the
13 of 22
American Institute of Aeronautics and Astronautics
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-8
-6
-4
-2
0
2
4
SA + Transition
SA, Fully Turbulent
Experiment
(a) r/R = 0.3
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-4
-3
-2
-1
0
1
2
(b) r/R = 0.47
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-6
-4
-2
0
2
(c) r/R = 0.80
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-6
-4
-2
0
2
(d) r/R = 0.95
Figure 16. Pressure coecient distributions for NREL wind turbine, V = 15m/s.
span, which appears as an interruption to the otherwise parallel chordwise ow across the span of the blade.
Skin friction in this transitional region is negative due to reversed ow adjacent to the blade surface.
The plots for the 10 m/s case, shown in Figs. 20(c) and 20(d), reveal that there is considerable ow
separation over the entire upper surface with the use of transition model. The turbulent simulation, on the
other hand, predicts that the separated region is conned to the trailing edge of the upper surface, where
the S809 separation ramp is located. Accordingly, the surface skin friction contours for the transitional case
reveal a feature-rich distribution of viscous stress which is missing in the fully turbulent case. It has to be
mentioned that the image shown is a single time snapshot of a highly unsteady ow, so the features shown
do not describe the uid eect on viscous forces completely.
At V

= 15 m/s, transition is predicted in the inboard region of the lower surface. The ow is fully
separated on the upper surface for the both simulations. However, the fully turbulent simulation shows a
small attached region towards the tip of the upper surface that contributes to discrepancy in the integrated
torque prediction described earlier in this section. For V

= 20 m/s, the ow is fully separated and there

are no sharp changes in skin friction indicative of transition.
Contours of eddy viscosity are shown in Figure 21. The eect of the transition model for the V

= 7 m/s
case, shown in Figure 21(a) and 21(b) are qualitatively similar to the 2D cases of Figure 4. Laminar ow is
expected in the front half of the airfoil, and in this region the eddy viscosity grows slowly when the transition
14 of 22
American Institute of Aeronautics and Astronautics
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-8
-6
-4
-2
0
2
SA + Transition
SA, Fully Turbulent
Experiment
(a) r/R = 0.3
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-4
-2
0
2
(b) r/R = 0.47
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-4
-2
0
2
(c) r/R = 0.80
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-2
-1
0
1
(d) r/R = 0.95
Figure 17. Pressure coecient distributions for NREL wind turbine, V = 20m/s.
model is applied. The fully turbulent simulation predicts a strong growth of eddy viscosity throughout the
surface of the airfoil and in the wake as a result. For V

= 10 m/s, shown in Figs. 21(c) and 21(d), the

transition model predicts a faster growth rate of eddy viscosity. As in the the 2D airfoil in Fig. 5, the
transition model is seen to generate lower levels of viscous damping and thus an earlier ow separation.
Contours at V

= 20 m/s are shown in Figures 21(e) and 21(f). Large separations are seen in these plots,
and the magnitude of eddy viscosity increases radically with increasing wind speed, leading to high levels of
mixing characteristic of turbulent ows.
V. Conclusions
In this work, the Re
t
SA transition model has been used in a RANS solver with overset grid
methodology for the analysis of ows over wind turbine airfoils and rotor blades. In situations for which ow
remains completely attached or massively separated, the transition model qualitatively and quantitatively
exhibits the same behavior as the fully turbulent model. When incipient and small separation regions are
present, the transition model is seen to signicantly improve the quality of the solution. This is seen in the 2D
validation of the S827 airfoil, where the transition model is able to capture the double slope of the lift curve
15 of 22
American Institute of Aeronautics and Astronautics
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-6
-4
-2
0
2
SA + Transition
SA, Fully Turbulent
Experiment
(a) r/R = 0.3
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-4
-3
-2
-1
0
1
(b) r/R = 0.47
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-3
-2
-1
0
1
2
(c) r/R = 0.80
x/c
C
p
0 0.2 0.4 0.6 0.8 1
-3
-2
-1
0
1
2
(d) r/R = 0.95
Figure 18. Pressure coecient distributions for NREL wind turbine, V = 25m/s.
and match the experimentally determined surface pressure coecient. The transition model also greatly
improves the accuracy of the 2D S809 simulation in the post-stall regime, another case that involves limited
ow separation. In the simulations of the NREL Phase VI wind turbine, the transition model substantially
improves the prediction of quantities at V

= 10 m/s. The continued need for improved modeling is evident

in the discrepancies in predicting the aerodynamic characteristics of the S827 airfoil in the middle range of
angles of attack as well as the considerable errors seen in the wind turbine simulations involving separation.
Accurate aerodynamic modeling of wind turbine ows remains a challenge in ow regimes that are dened
by transition and separation. As LES and DNS will continue to be infeasible for industrial applications in
the next decade, accurate RANS based models will remain desirable for these problems. Further develop-
ment of these models requires more highly rened experimental datasets as well as DNS results in simpler,
isolated situations, such that the eect of rotation, secondary ows and turbulence anisotropy can be better
characterized, leading to improved predictions of transition and separation.
16 of 22
American Institute of Aeronautics and Astronautics
(a) V = 7m/s, without transition (b) V = 7m/s, with transition
(c) V = 10m/s, without transition (d) V = 10m/s, with transition
(e) V = 15m/s, without transition (f) V = 15m/s, with transition
(g) V = 20m/s, without transition (h) V = 20m/s, with transition
Figure 19. Surface pressure contours for NREL wind turbine.
17 of 22
American Institute of Aeronautics and Astronautics
(a) V = 7m/s, without transition (b) V = 7m/s, with transition
(c) V = 10m/s, without transition (d) V = 10m/s, with transition
(e) V = 15m/s, without transition (f) V = 15m/s, with transition
(g) V = 20m/s, without transition (h) V = 20m/s, with transition
Figure 20. Contours of skin friction along with surface streamlines for NREL wind turbine.
18 of 22
American Institute of Aeronautics and Astronautics
(a) V = 7m/s, without transition (b) V = 7m/s, with transition
(c) V = 10m/s, without transition (d) V = 10m/s, with transition
(e) V = 20m/s, without transition (f) V = 20m/s, with transition
Figure 21. Sectional eddy viscosity contours.
19 of 22
American Institute of Aeronautics and Astronautics
VI. Acknowledgement
This work is partially supported by the DoD NDSEG fellowship and by the DoE ASCR program on
Fluid/Structure interactions in Wind Turbine Applications at Stanford University.
A. Description of the Re
t
transition model
The original Re
t
transition model was developed by Langtry et al.
19
for use with the SST-k
turbulence model. One advantage of this model over many other transition models is that Re
t
does
not require the integration of a boundary layer followed by a search for critical Re

at which transition
onset begins. Furthermore, because this model allows intermittency to vary across the boundary layer, it
is able to capture transition triggered by a laminar separation bubble without need for further correction.
This is particularly advantageous in low speed ows, where separation bubbles are frequently the cause of
transition. The Re
t
model is correlation-based and provides a convenient framework wherein users may
insert proprietary or internal correlations.
The Re
t
SA model, introduced by Medida et al.
22
adapts this method to work with the one equation
Spalart-Allmaras turbulence model. Because the Re
t
SA model has been introduced recently, and
because its implementation in OVERTURNS is original, the present work performs additional validation to
conrm its credibility in transitional ows such as those seen in wind turbine applications.
In addition to the RANS equation and turbulence model equation, the Re
t
SA model requires
the solution of two transport equations. The rst is for intermittency, , a quantity which varies from zero
to one and represents the probability of turbulent ow at a given location. The second is for the transition
momentum thickness Reynolds number Re
t
, the purpose of which is to convect the eects of turbulence
intensity from the freestream into the boundary layer.
When coupled with Re
t
SA, the equation for the Spalart-Allmaras variable becomes
D
Dt
=

P

+
1

_
(( + ) ) +c
b2
( )
2
_
(5)
where the source terms

P

and

D

are modied from the original turbulence model to depend on inter-

mittency as follows:

P =
eff
P

,

D

= max(min(, ), 1.0)D

, = 0.5 (6)
The model parameter
eff
will be dened below. The original source terms P

and D

and remaining
closure constants are those given by the Spalart-Allmaras model:

t
=
t
f
1
, f
1
=

3

3
+c
3
1
, =

(7)
P

= c
b1

and D

= c
w1
f
w
_

d
_
2
(8)

= +

2
d
2
f
2
, f
2
= 1

1 +f
1
(9)
f
w
= g
_
1 +c
6
w3
g
6
+c
6
w3
_
1
6
, g = r +c
w2
(r
6
r), r =

2
d
2
(10)
c
b1
= 0.1355, =
2
3
, = 0.41, c
w1
=
c
b1

2
+
1 +c
b2

, c
w2
= 2.0, c
1
= 7.1 (11)
The transport equation for intermittency is
D()
Dt
= P

+

x
j
__
+

t

f
_

x
j
_
(12)
with auxillary equations given by
P

= F
length
c
a1
S[F
onset
]
0.5
(1.0 c
e1
) (13)
20 of 22
American Institute of Aeronautics and Astronautics
D

= c
a2
F
turb
(c
e2
1.0) (14)
F
onset
= max(F
onset2
F
onset3
, 0) (15)
F
onset1
=
Re

2.193Re
c
(16)
F
onset2
= min(max(F
onset1
, F
4
onset1
), 2.0) (17)
F
onset3
= max
_
1
_
R
T
2.5
_
3
, 0
_
(18)
F
turb
= e

R
T
4

4
(19)
Re

=
d
2
S

, Re
T
=

t

(20)
The transport equation for critical momentum thickness Reynolds number is
D(Re
t
)
Dt
= P
t
+

x
j
_

t
( +
t
)
Re
t
x
j
_
(21)
with auxillary relations given by
P
t
= c
t

t
(Re
t
Re
t
)(1.0 F
t
) (22)
F
t
= min
_
max
_
F
wake
e
(
d

)
4
, 1.0
_
1/c
e2
1.0 1/c
e2
_
2
_
, 1.0
_
(23)
A set of correlations for critical momentum thickness Reynolds number is:
Re
t
=

(1173.51 589.428Tu +
0.2196
Tu
2
)F(

), Tu 1.3
331.50[Tu 0.5658]
0.671
F(

), Tu > 1.3
(24)
F(

) =

1 [12.986

123.66
2

405.689
3

] e
[
Tu
1.5
]
1.5
,

0
1 + 0.275[1 e
[35.0

]
]e
[
Tu
0.5
]

> 0
(25)

=

2

dU
ds
(26)

BL
=
Re
t

U
;
BL
= 7.5
BL
; =
50d
U

BL
(27)
and the model constants are
c
e1
= 1.0; c
q1
= 2.0; c
e2
= 50.0; c
a2
= 0.06;
f
= 1.0 (28)
c
t
= 1.0;
t
= 2.0 (29)
To improve the prediction of transition due to laminar separation bubble, the following correction is used.

sep
= min
_
s
1
max
_
0,
_
Re

3.235Re
c
_
1
_
F
reattach
, 2.0
_
F
t
(30)

e
= max(,
sep
) (31)
F
reattach
= e

R
T
20

4
, and s
1
= 2.0 (32)
21 of 22
American Institute of Aeronautics and Astronautics
References
1
U.S. Department of Energy, 20% Wind Energy by 2030, Increasing Wind Energys Contribution to U.S. Electricity
Supply, DOE/GO-102008-2567, 2008.
2
M. O. Hansen, Aerodynamics of Wind Turbines, 2nd Ed., Earthscan, 2008.
3
J. G. Leishman, Challenges in Modelling the Unsteady Aerodynamics of Wind Turbines, Wind Energy, Vol. 5, No. 2,
pp. 85132, 2002.
4
D. Simms, S. Schreck, M. Hand, and L.J. Fingersh, NREL Unsteady Aerodynamics Experiment in the NASA-Ames
Wind Tunnel: A Comparison of Predictions to Measurements, NREL/TP-500-29494, 2001.
5
M. M. Hand, D. A. Simms, L. J. Fingersh, D. W. Jager, J. R. Cotrell, S. Schreck, and S. M. Lawood, Unsteady
Aerodynamics Experiment Phase VI: Wind Tunnel Test Congurations and Available Data Campaigns, NREL/TP-500-29955,
2001.
6
N. N. Srensen, J. A. Michelsen, and S. Schreck, Navier-Stokes Predictions of the NREL Phase VI Rotor in the NASA
Ames 80 ft 120 ft Wind Tunnel, Wind Energy, Vol. 5, No. 2, pp. 151169 2002.
7
E. P. N. Duque, M. D. Burklund, and W. Johnson, Navier-Stokes and Comprehensive Analysis Performance Predictions
of the NREL Phase VI Experiment, Journal of Solar Energy Engineering, Vol. 125, No. 4, 2003.
8
M. A. Potsdam and D. J. Mavriplis, Unstructured Mesh CFD Aerodynamic analysis of the NREL Phase VI Rotor,
AIAA Aerospace Sciences Meeting, Orlando, FL, January 2009.
9
P. Devinant, T. Laverne, and J. Hureau, Experimental study of wind-turbine airfoil aerodynamics in high turbulence,
Journal of Wind Engineering and Industrial Aerodynamics, Vol. 90, No. 6, pp. 689707, June 2002.
10
R. Michel, Etude de la Transition sur les Proles dAile, ONERA Report 1/1578A, 1951.
11
J. Moran, An Introduction to Theoretical and Computational Aerodynamics, Dover Publications, Mineola, New York,
1984.
12
R. Eppler, A Computer Program for the Design and Analysis of Low-Speed Airols, NASA Langley Research Center,
NASA-TM-80210, N80-29254, 1980.
13
D. M. Somers, Design and Experimental Results for the S809 Airfoil. NREL/SR-440-6918, Jan. 1997.
14
D. M. Somers, Design and Experimental Results for the S827 Airfoil. Airfoils, Inc., 1999.
15
R. J. A. Howard, M. Alam, and N. D. Sandham, Two-Equation Turbulence Modelling of a Transitional Separation
Bubble, Flow, Turbulence and Combustion, Vol. 63, 1999.
16
A. M. O. Smith and N. Gamberoni, Transition Pressure Gradient and Stability Theory, Douglas Aircraft Company,
Long Beach, California, Rep. ES 26388, 1956.
17
J. L. Van-Ingen, A Suggested Semi-Empirical Method for the Calculation of the Boundary Layer Transition Region,
Univ. of Deft, Dept. Aerospace Engineering, Rep. VTH-74, 1956.
18
D. D. Pasquale, A. Rona, and S. J. Garrett, A Selective Review of CFD Transition Models, 39
th
AIAA Fluid Dynamics
Conference, San Antonio, TX, June 2009.
19
R. B. Langtry and F. R. Menter, Transition Modeling for General CFD Applications in Aeronautics, 43rd AIAA
Aerospace Sciences Meeting and Exhibit, Reno, Nevada, January 2005.
20
R. B. Langtry, J. Gola, and F. R. Menter, Predicting 2D Airfoil and 3D Wind Turbine Rotor Performance using a
Transition Model for General CFD Codes, 44th AIAA Aerospace Sciences Meeting and Exhibit, Reno, Nevada, January 2006.
21
F. R. Menter, Two-Equation Eddy-Viscosity Turbulence Models for Engineering Application, AIAA Journal, Vol. 32,
No. 8, pp. 15981605, 1994.
22
S. Medida and J. Baeder, Numerical Prediction of Static and Dynamic Stall Phenomena using the Re
t
Transition
Model, American Helicopter Society 67
th
Annual Forum, Virginia Beach, VA, May 2011.
23
P. R. Spalart and S. R. Allmaras, A One-equation Turbulence Model for Aerodynamic Flows, AIAA Paper 1992-0439,
30th AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, January 69, 1992.
24
V. K. Lakshminarayan, Computational Investigation of Micro-Scale Coaxial Rotor Aerodynamics in Hover, Ph.D.
dissertation, Department of Aerospace Engineering, University of Maryland at College Park, 2009.
25
P. E. O. Buelow, D. A. Schwer, J. Feng, C. L. Merkle, A Preconditioned Dual-Time, Diagonalized ADI scheme for
Unsteady Computations, 13th AIAA Computational Fluid Dynamics Conference, Snowmass Village, CO, June 1997.
26
S. A. Pandya, S. Venkateswaran, and T. H. Pulliam, Implementation of Preconditioned Dual-Time Procedures in
OVERFLOW, 41st AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV, January 2003.
27
T. Pulliam, and D. Chaussee, A Diagonal Form of an Implicit Approximate Factorization Algorithm, Journal of
Computational Physics, Vol. 39, No. 2, pp. 347363, 1981.
28
E. Turkel, Preconditioning Techniques in Computational Fluid Dynamics, Annual Review of Fluid Mechanics, Vol. 31,
January 1999, pp. 385416.
29
Y. Lee On Overset Grids Connectivity and Vortex Tracking in Rotorcraft CFD, PhD Dissertation, University of Maryland,
College Park, 2008.
22 of 22
American Institute of Aeronautics and Astronautics